Configuration for TensorFlow Agents PPO on MinitaurBulletEnv-v0

tensorflow-agents-ppo-minitaur.py

def minitaur_config():
  # General
  algorithm = ppo.PPOAlgorithm
  num_agents = 10
  eval_episodes = 30
  use_gpu = False
  # Environment
  env = 'MinitaurBulletEnv-v0'
  max_length = 1000
  steps = 1e7  # 10M
  # Network
  network = networks.feed_forward_gaussian
  weight_summaries = dict(
      all=r'.*', policy=r'.*/policy/.*', value=r'.*/value/.*')
  policy_layers = 200, 100
  value_layers = 200, 100
  init_mean_factor = 0.3
  init_logstd = -1
  # Optimization
  update_every = 30
  policy_optimizer = 'AdamOptimizer'
  value_optimizer = 'AdamOptimizer'
  update_epochs_policy = 25
  update_epochs_value = 25
  policy_lr = 1e-4
  value_lr = 3e-4
  # Losses
  discount = 0.995  # Important.
  kl_target = 1e-2
  kl_cutoff_factor = 2
  kl_cutoff_coef = 1000
  kl_init_penalty = 1
  return locals()